Expertise conundrum (wk 28/2025)

The AI expertise conundrum; strong winds and forest fires, frosty fingers, household goods, tuna collar; animal chiropractors, making clothing in various places, exceptional R&D contractors.

Hello friends,

An announcement: idk, the first (and frankly, the only) training tool for productive discomfort, is back in stock for US shipping addresses.

Early this week, strong winds turned a small automobile fire on the autoroute into a wildfire that consumed 750 hectares of forest just north of Marseille. We were en route to the Marche aux Puces when the fire really got going. Wood ash blew into the poky Maghrebi traiteur we ended up in, and onto lunch (stuffed peppers).

A forest fire north of Marseille, as seen from the A55; 8 July 2025.

Apart from that, the South finally cooled down during the week. It became possible once more to do things like write without thinking only of frosty drinks and sleep well at night. Some nights got cold enough that the sea surface temperature dropped enough that I got frosty fingers after a 30-minute swim at Catalans on Thursday.

On Friday, I went up to the Luberon to see some friends who have run an opinionated select shop for decades. It is very restful to visit a house in which nearly every object that can be seen and used has been chosen to be well-made (not the same thing as expensive). Equipping and furnishing a house can be a multi-year series of meaning-making acts.

In the Luberon; 12 July, 2025.

Then, perhaps unwisely, I picked up a tuna collar at the Vieux-Port on Sunday, and spent far too long breaking it down incompetently and without supervision.

Writing

Over the last few months, I’ve had increasingly frequent fascinating conversations with different people about the possibility of a different model of human-machine interaction that accommodates the particular seductions of AI systems. Sounds theoretical, but actually has immediate practical implications on how to build AI tools. Anyway, I wrote something this week about the AI expertise conundrum:

Current LLMs can’t truly or create new knowledge on their own; but they can help humans do that innovation work more quickly. LLMs work best as eager but inexperienced research assistants: good at mapping known landscapes, bad at deciding what matters. So, paradoxically, they’re most useful to people with enough domain expertise to ask good questions and spot flaws — but they leave novices vulnerable to plausible but biased or simply incorrect outputs. If your organisation is deploying AI as a creativity engine or innovation driver … maybe reconsider? The smarter approach: Design AI use around its real affordances (information synthesis, not autonomous creation) and build AI workflows that keep human meaning-making at the centre.

👉  Read more here about the AI expertise conundrum. 

In a garden supply centre on Rue Paradis; 10 July, 2025.

Elsewhere

1. On farm animal chiropractors: ‘“This isn’t a replacement for traditional veterinary medicine,” said Stief. “But I will say I think the veterinary field — just like human medicine has — is getting more open-minded and holistic in its approach. I’ve never had a veterinarian respond poorly to the work I do … An animal that feels better in their body is going to have lower cortisol levels, it’s going to feel better, it’s going to digest its food better and so what you’ll notice is an animal — whether it’s a farm animal or a horse — is that if it feels better, you’re going to see overall improvement in the quality of life, whether you’re measuring performance or digestion, you’re going to see changes as that body falls back into homeostasis and balance.”’

2. On making clothing in the US (and elsewhere): “It's not just custom tailoring. South Korea is unique in that it's full of small, independent brands that are doing cool things — Frizmworks, Document, Eastlogue, etc. These brands are unusual in that they offer distinct designs at relatively affordable prices … South Korean factories are happy to work with smaller brands, who only require small runs. The wages are also lower than many parts of the world. In China, wages are relatively low, but factories require big runs … The combination of low wages + small minimums means that tiny, independent brands can say "hey make this weird robe coat or military parka that I think only level 100 menswear nerds will buy." The item ends up being unique enough to be interesting, but affordably priced to sell.”

3. On the value of exceptional R&D contractors: “Exceptional contractors … shared three distinct traits, rarely found together. [Each exceptional contractor:]

   1. Was novelty-seeking, with a strong preference for projects that pushed the technological frontier forward substantially.

   2. Built useful technology for actual users. This entailed professional contract management and a willingness to focus on difficult systems engineering tasks.

   3. Used more flexible team structures than academia. When compared to academia, they more effectively hired, organized, and incentivized researchers, engineers, and other experts to collaborate on applied projects in a common-sense fashion.”

See you next week,
VT

Palais du Pharo from Mucem; 6 July, 2025.

Fallacy lens (wk 27/2025)

Optimisation as fallacy; business as meaning-making act; the drudgery of administration; the search for good manufacture; working in a canicule.

Hello friends,

This week, France has turned to a leathery and exhausted crisp while gripped by a monstrous Europe-wide heat wave. I’m now back in France until July 26 to deliver a few pieces of work and continue the apparently unending process of fixing up the apartment — will be between Ljubljana (15-18), Paris (21-24), and Marseille (the rest of the time). Drop me a line if you’re in any of those places and want to get a coffee or, supply permitting, a glass of wine.

Midweek, I started gathering tax documentation for the small consulting company I work through. It’s already taken over 15 hours of sorting through three email accounts and folders of receipt photos, invoices, bank statements, and other miscellaneous digital bits and pieces. I’ve not even started on the paper receipts that I put into a series of boxes to deal with later (= now).

Is this is the moment I finally move to using some kind of application or service that will a) send invoices in multiple currencies with live historical conversion rates, b) do automatic bookkeeping and reimbursements from machine-processed receipt submissions, all at a price reasonable for a business with a two-person headcount (for now). Do you have a solution you’re happy with? Drop me a line as this inquiring mind would like to know.

At Catalans (2 July, 2025).

Writing

I wrote two short pieces this week.

As many of you probably know, I’ve spent nearly two decades trying to unpack the surprisingly textured and poorly understood concept of uncertainty, and develop some kind of useful set of practices for dealing with different types of non-risk uncertainty. Understanding the difference between uncertainty and risk is central to being able to respond properly to an increasingly uncertain, not simply risky, world.

A futurist friend recently made an expensive, early exit from the Middle East during a flare-up in the Israel-Iran conflict ... then felt embarrassed when the situation de-escalated. That embarrassment stems from misframing the decision as an optimisation problem — which assumes that precise estimation of risks, probabilities, and timings is possible. It wasn’t. The situation wasn’t only risky, it was also uncertain — marked by capricious actors uninterested in playing by well-understood rules. In uncertain situations, optimisation fails and other frames for decisionmaking make much more sense. The real skill here is correctly diagnosing the type of not-knowing you’re facing, then choosing a reasoned and appropriate decision frame instead of defaulting to a decision frame only suited to risk.

👉 Read more here about why optimisation is the wrong decision frame for uncertain situations.

In the last few years, I’ve developed a language around four different types of not-knowing that aren’t risky. The most important type is also the least understood and discussed: Not-knowing about relative value. This is the form of uncertainty that underlies the writing I’ve been doing over the last 18 months on meaning-making as a lens for considering AI.

Anthropic’s experiment with using Claude as an autonomous shopkeeper agent failed — not just because the AI was gullible, but because running even a simple business involves inherently human, meaning-making decisions. Doing business isn’t just executing tasks like pricing or inventory. It’s deciding what matters, what to trade off, and what success looks like. These are subjective choices without correct answers. Until AI systems can make meaning, they shouldn’t be tasked with running businesses on their own. The real question isn’t whether an AI is gullible, but whether the work requires meaning-making. If it does, that work must remain human.

👉  Read more here about why doing business is a meaning-making act. 

Morning facing the Mediterranean (4 July, 2025).

Elsewhere

Lufthansa seems to have lost one of my bags, prompting a low-level search for replacement clothing for next week. Can I really be the only person who wants a small collection of relatively basic normal-person clothing that a) isn’t technical clothing, workwear, or athleisure, b) is made not by a MegaCorp, c) uses well-sourced material and fabrication techniques, c) has interesting but not outlandish cuts that are actually wearable in a wide range of settings not just at a vernissage, d) is designed around normal washing instead of drycleaning or other forms of specialist clothing care. Why is it so hard to find such things outside of bespoke?

I canvassed the small number of clothing industry experts I know. Below are some of the recommendations that came back which seemed especially good to me, but only a few met the full brief — especially, workwear seems nearly inescapable.

1. Casatlantic

2. James Coward

3. Old Town

4. S.EH. Kelly

5. Oliver Church

6. Evan Kinori

7. Anglo-Italian

8. mfpen

9. Auralee

10. Front Office

11. Drake’s

12. Sunspel

And the consensus opinion (mine included) is that some items from MegaCorps are okay because the alternatives from non-MegaCorps are better but not enough to justify the massive price difference (e.g., t-shirts from James Perse). Drop me a line if you know of a maker I should look at.

Well-used cloth (5 July, 2025).

See you next week,
VT

Tool-building (wk 26/2025)

Uncomfortable questions about unknowns; building AI tools for better meaning-making; problems, learned helplessness, performance, comfort, and dissonance.

Hello friends,

This past week, I was on several long videoconferences with senior management of an organisation I’m working with. During each meeting, the discussion predictably circled the growing number of unknowns that are messing up their global business planning and strategy. Also predictably, the conclusion was to deal with these unknowns by investing more energy into cataloguing them and quantifying their probabilities.

In a tough spot (one in a series of postcards in Heman Chong’s installation, “Perimeter Walk”; 29 June, 2025).

Climate change and extreme weather, upheavals in economies and labour markets, political transition, geostrategic conflict, natural hazards, unexpected side effects of financialisation, novel disease, new technology. These disruptions aren’t stopping, and they increasingly compound each other. The unknowns causing these disruptions are truly uncertain one, not just risky. Is the right approach to dealing with unknowns to just pour time, effort, and money into trying (and inevitably failing) to quantify them accurately?

The organisation I’m working with is clearly not alone. I’m continually being sent articles about how uncertainty is penetrating more domains of life, that uncertainty is increasingly unavoidable and becoming more intense, that leaders must become sophisticated managers of uncertainty, etc. It’s good that more journalists and thinkers are recognising the difference between risk and uncertainty. But where these articles offer suggestions about how to respond to this onrushing wave of uncertainty, the recommendations are always old wine, and not even in new bottles. The advice conflates uncertainty and risk, uses a risk mindset to approach true uncertainty, and doesn’t fundamentally rethink the risk-framed outcomes we choose to pursue. Will using these comfortable risk-oriented ways of thinking lead us in the wrong direction in a world increasingly dominated by uncertainty?

At the same time, without a more sophisticated way to understand uncertainties that aren’t simply risky, we can’t take full advantage of the opportunities for novelty, innovation, and change that uncertainty creates space for. If uncertainty is increasingly intense, unavoidable, and ubiquitous, shouldn’t we be finding better ways to understand its potential for generativity?

Asking these questions about unknowns is uncomfortable because they lead to uncomfortable conclusions about how unready we are for an uncertain (not just risky) world. The instinctive reaction to uncomfortable thoughts is to resort to comfortable old thinking. Isn’t it time for that to change?

Behind an art moving facility (29 June, 2025).

Writing

A startup called Cluely has been marketing its AI tool by promising that “today is the start of a world where you’ll never have to think again.” This is a dire prospect. I’ve been working on a project to figure out what education and educational tools should be in a world where AI is increasingly ubiquitous. It’s still not ready for public consumption but I’ve written another short piece on building AI tools for better meaning-making. tl;dr: Instead of building tools that are designed to generate outputs indistinguishable from human outputs, we should be building AI tools that focus on helping users learn to do meaning-making: The work of making inherently subjective decisions about the subjective and relative value of things. You can read the whole thing here.

Tools at Domaine de l’Arbre Blanc (25 September, 2017).

Elsewhere

As I continue working through practical questions about structuring user interactions to make meaning-making explicit, my old files have been more helpful than I’d expected — though not in ways that I’d have expected.

Tom West, skunkworks impresario at Data General, in Tracy Kidder’s The Soul of a New Machine:

I traveled with West to New York. We stopped at a grocery store in which the cash registers were equipped with one of those devices that reads the price of an item automatically, a computerized checkout system. This one wasn't working well. West got down on his hands and knees and poked his head in under the cashier's counter to have a look at the thing. The clerk made her mouth an O. When West came out, dusting off his hands, he explained that he had helped design this particular model when he had worked at RCA. "It's a kludge," he said grinning. The clerk had some trouble figuring what the beer we bought ought to cost, and as we left, West said, out of her earshot, "Ummmmh, one of the problems with machines like that. You end up making people so dumb they can't figure out how many six-packs are in a case of beer."

Chuck Close, in Joe Fig’s Inside the Painter’s Studio:

I think our whole society is much too problem-solving oriented. It is far more interesting to [participate in] ‘problem creation’ … You know, ask yourself an interesting enough question and your attempt to find a tailor-made solution to that question will push you to a place where, pretty soon, you’ll find yourself all by your lonesome — which I think is a more interesting place to be.

Daniel Lieberman, in an edge.org interview (also and incidentally a fine endorsement for investing in productive discomfort):

We love comfort, and people make a lot of money selling us comfort, but I would challenge the notion that comfort is usually good for us.

William Carlos Williams, from Paterson IV:

Dissonance

(if you are interested),
leads to discovery

See you next week,
VT

Pineapples where they shouldn’t be (26 June, 2025).

Before and after science (wk 25/2025)

A workshop on meaning-making; the future of education is meaning-making; question-value, the aesthetic value of uncertainty; what happens before and after science.

Hello friends,

This week, I ran a half-day workshop in Adelaide on different types of not-knowing that aren’t risk. The group was multi-disciplinary (psychologists, educators, mathematicians, curators, designers, futurists, artists, social workers, and statisticians) and I’m told that the participants enjoyed it and found it insightful and useful.

Hard at work on meaning-making (20 June, 2025).

I’ll write more on this in the next few weeks but my main takeaways are:

1. It takes at least an hour to effectively explain the different types of not-knowing, but the value and relevance of the construct and typology is clear and seems applicable across a wide range of disciplines,

2. It takes so long because “risk” and “uncertainty” have been used imprecisely and inconsistently to describe not-knowing for decades (even by people who should know better) and this obstructive underbrush has to be cleared before a new foundation for understanding can be built,

3. The framing of not-knowing as a generative principle is what gets people to really engage with the typology of not-knowings, and

4. People with different disciplinary backgrounds come up with unexpected (to me) ways to experience the different types of not-knowing, many of which can be implemented quickly and experienced in minutes (vs hours or days).

Writing

All through June I’ve been working on a project which isn’t ready for the public yet — it’s about what education and educational tools look like in a world where AI is increasingly ubiquitous. I’ve written something short about it for now. This is the tl;dr version: AI tools are increasingly accessible, cheap, and seem potentially able to produce any output a human can produce — they’ll certainly reconfigure what work looks like. I agree with this read on AI, except for one important difference: Humans can and must do what I call “meaning-making” because AI can’t do it yet. Meaning-making consists of making subjective judgments about the relative value of things. Education at all levels, but especially higher education, has largely abandoned teaching students how to make and justify subjective value judgments. To remain relevant, education must reorient around helping students learn what meaning-making is, and how to do it well.

You can read the whole thing, plus more context at the links below.
👉 The future of education is meaning-making.
👉 What makes us human for now. [context]
👉 The meaning-making lens on AI. [more context]

Marino Rocks (20 June, 2025).

Elsewhere

This week I’ve been thinking about partial knowledge a lot, and going through my old files (in some cases from decades ago) to see if I could find more ways to talk about the benefits of partial knowledge.

Robert Irwin; in Lawrence Weschler, Seeing is forgetting the name of the thing one sees:

I find it very precarious for a culture only to be able to measure performance and never to be able to credit the questions themselves ... As the questions go up, the performance level goes down — and that's natural, since people don't yet know how to act on those questions, they're stumbling around in a fog — whereas when performance goes up the quality of the questions tends to go down.

Donald Francis Tovey, Essays in musical analysis, vol. 3:

Theorists are apt to vex themselves with vain efforts to remove uncertainty just where it has a high aesthetic value.

Philipp Otto Runge to Johann Wolfgang Goethe (3 July, 1806), in Goethe’s Zur Farbenlehre (1810):

There have been those who have built bridges and suspension work and other such technical things simply by eye. That is certainly possible for a time, but once a certain height has been reached and one naturally hits upon mathematical conclusions, his whole talent will be for nothing unless he works his way through the science and back into freedom.

See you next week,
VT

Someone else’s music (21 June, 2025).

Severely overinsulated (wk 24/2025)

Hot air balloon conferences; ideal-typical convenings; open space; delicious trash foods in South Australia.

Hello friends,

This morning, I ran a half-day workshop on distinguishing uncertainty from risk, and on using signals and scenarios as a strategy tool when facing uncertainty. It was for a mixed group from risk, resilience, strategy, and continuity functions at SA Water, the state water utility of South Australia. If you want something like that done for your team or organisation, you could drop me a line.

The Botanic Gardens, Singapore (8 June, 2025).

Writing

In the last decade, I’ve designed and facilitated several convenings and been a participant at many more. The ones which were the hardest to do (and to get approved) were also the ones most worth doing and being a part of. What do I mean? Most convenings are what I call Your Usual Conference, built around a “fixed schedule of keynotes, papers, and panels published in advance.” These are rarely easy to actually run, but they are easy to conceptualise and conceptually easy to plan. So the people paying for them and running for them know what to expect — Your Usual Conference is easy to get approved and to actually do. But they’re never Great.

Curiosity Camp 2024, Cazadero, CA (2 June 2024).

For a convening to rise to the level of being Great means deep memorability, unexpected but strong connections formed, and transformational exposure to new perspectives and ways of thinking. The first MAD Symposium in 2011 was Great, so was MAD5 in 2016, Science Foo Camp in 2008, the Bar Camps in Boston in 2011/12, Kinnernet 2013, and Curiosity Camp 2024. For Greatness to happen, the structure of the convening must respond to what is happening during the convening. A predesigned conference is definitionally unresponsive. In its ideal form, a Great convening is an emergent phenomenon, and it always requires extremely good facilitation, an absurd amount of preparation, and an unrealistic willingness to not-know what is going to happen. I love Great Convenings, and it turns out I’ve written about aspects of this for some time:
👉 Behind the scenes at a non-linear, emergent conference.
👉 The ideal form of a convening (for me).
👉 Why designing for open-endedness is hard but essential.

Setting up for MAD5, Copenhagen (24 August, 2016).

Elsewhere

I’ve been in Adelaide at MOD. for a week, and have already had a chicken parm, some Cherry Ripe, and a bunch of Tim-Tams. Still looking for FruChocs, a burger with the lot, a good frog cake, and the infelicitously named Chico Babies. All the fancy people are wearing RM Williams, and all the tradies are wearing Blundstones. The central business district is appealingly small, walkable, and heterogeneous — it feels more like the core of Marseille than the City of London. The weather right now, in the Depths of Winter, is appropriate for T-shirts. I was told by Locals to bring lots of warm clothing so am severely overinsulated. Soon, I hope to report favourably on my attempts to obtain a secondary market NST egg.

The Botanic Gardens, Adelaide (10 June, 2025).

See you next week for real,
VT